agglomeration, backward and forward linkages: evidence from … · agglomeration, backward and...

Agglomeration, backward and forwardlinkages: evidence from South Koreaninvestment in China

Peter Debaere Darden Business School, University of VirginiaJoonhyung Lee Economics Department, University of Texas

at AustinMyungho Paik School of Law, University of Texas at Austin

Abstract. With a firm-level data set, we study the location decision of South Koreanmultinationals across China’s regions. Our conditional logit estimates confirm agglomer-ation effects along industry and along national lines. We add an upstream and downstream(backward and forward) linkage effect. We find that the presence of upstream and down-stream South Korean affiliates significantly increases the likelihood that a South Koreanmultinational invests in a particular region. However, linkages that do not differentiateby nationality do not seem to matter much. As such, our analysis of investors’ locationchoice brings together two perspectives: linkages and agglomeration along national lines.JEL classification: F21, F23, R12

Agglomeration, liens en amont et liens en aval : resultats pour l’investissement sud coreen enChine. A partir d’une base de donnees au niveau de l’entreprise, on etudie la localisationde firmes plurinationales sud-coreennes dans les regions de la Chine. Les resultats d’uneanalyse de type logit confirment qu’il existe des effets d’agglomeration tant dans l’axeindustriel que national. Quand on ajoute les effets en amont et en aval, on decouvre quela presence d’affilies sud-coreens en amont et en aval accroıt de maniere significative laprobabilite qu’une firme plurinationale sud-coreenne investisse dans une region partic-uliere. Cependant les liens qui ne discriminent pas selon l’axe de la nationalite ne semblentpas avoir grande importance. L’analyse des decisions de localisation d’investissementsynthetise deux perspectives : liens et agglomeration selon l’axe de la nationalite.

1. Introduction

The distribution of South Korean affiliates across Chinese provinces is veryuneven and is also more concentrated than the overall distribution of foreign

Debaere is also affilitated with CEPR. An earlier version of the paper benefited frompresentations at the University of Nottingham, the Texas Econometric Camp, and the DardenBusiness School of University of Virginia. The usual disclaimers apply. Email:[email protected]; [email protected]; [email protected]

Canadian Journal of Economics / Revue canadienne d’Economique, Vol. 43, No. 2May / mai 2010. Printed in Canada / Imprime au Canada

0008-4085 / 10 / 520–546 / C© Canadian Economics Association

Agglomeration, backward and forward linkages 521

affiliates in China. As figures 1 and 2 indicate, four regions out of 26 coverroughly about 75% of South Korean affiliates in China. Moreover, this pro-nounced pattern is not limited to China. As a matter of fact, if you turn tothe U.S., another large country with many regions, you find that almost 60% ofSouth Korea’s affiliates are concentrated in one state: California. In this paperwe investigate these patterns of clustering of multinationals along national linesfor South Korean affiliates in China, and we want to better understand whatdetermines this type of agglomeration. That affiliates of multinationals clusteralong national lines is a well-known phenomenon that has also been observed inother countries. It has triggered empirical work, among others, by Head, Ries,and Swenson (1995) that has been particularly influential. This type of work issuggestive of the impact that agglomeration externalities along national lines mayhave on a firm’s decision to locate in a particular region. While the evidence ofagglomeration is abundant, however, it is not always clear what exactly is drivingthe agglomeration. In this paper, we investigate the role that the availability ofbackward and forward linkages can play in location decisions.

Important for the approach that we take is the work of Amiti and Javorcik(2008). Amiti and Javorcik find that forward and backward linkages are animportant determining factor for multinationals from any country to invest inone of China’s regions. Building on an economic geography model, they findthat foreign firms choose to locate in the regions where they can easily supplytheir intermediate goods to others or purchase intermediate goods from otherfirms.1 To operationalize the upstream and downstream links, the authors useinput-output tables that were advocated long ago by Hirschman (1958) and theyinteract these with total (local and foreign) regional industry outputs. In theiranalysis, Amiti and Javorcik (2008) thus focus on total (local and foreign) industryoutput and do not separately consider the linkages through multinationals.2

In our study of South Korean investment decisions across Chinese regions,we combine the approach of Head, Ries, and Swenson (1995) and Amiti andJavorcik (2008). We let the presence of South Korean firms in China affect thelocation decision of South Korean multinationals in three ways. In addition tothe regular agglomeration effect that is captured by the number of South Koreanaffiliates in nearby industries, we use input-output tables to investigate the extentto which the presence of South Korean upstream or downstream affiliates innearby industries increases the probability that a South Korean multinational willinvest in a particular Chinese region. At the same time, we explicitly compare andcontrast these linkage effects with the ones from the total number of upstreamand downstream companies in an industry irrespective of their nationalities. In

1 Gorg and Strobl (2002) also study agglomeration effects and linkages in Ireland. FollowingMarkusen and Venables (1999), they investigate the extent to which the presence ofmultinationals affects entry of indigenous firms in Ireland and how the location choice isaffected by downstream linkages.

2 In robustness checks, Amiti and Javorcik (2008) find that the number of foreign firms in a regionalso plays a role, in addition to the total linkage effects.

0 -

5%

5 -

10%

10 -

25%

25 -

100%

FIG

UR

E1

Dis

trib

utio

nof

Sout

hK

orea

nin

vest

men

tin

Chi

naSO

UR

CE

:Exp

ort-

Impo

rtB

ank

ofK

orea


010

20

30

40

Perc

enta

ge o

f W

orl

d/K

ore

an a

ffili

ate

s

Anhui

Beiji

ng

Fujia

n

Ga

nsu

Guangdong

Guiz

hou

Hain

an

Heib

ei

Heilo

ngjia

ng

He

na

n

Hubei

Hu

na

n

Jia

ngsu

Jia

ngxi

Jili

n

Lia

onin

g

Qia

nghai

Shaanxi

Shangdong

Shanghai

Shanxi

Sic

huan

Tia

njin

Yu

nn

an

Zhejia

ng

World Korea

FIGURE 2 Distribution of multinationals in ChinaSOURCE: For worldwide multinational firms: China Statistics Yearbook 2006. For Korea:Export-Import Bank of Korea

doing so, we insert a national dimension into the analysis of Amiti and Javorcik(2008) and a linkages dimension into the work of Head et al. (1995). Interestingly,we find that linkages along national lines matter most for South Korean investors.However, the linkages at the industry level that do not differentiate by nationalitydo not play much of a role.

FDI has been among the fastest growing international indicators and itsgrowth toward China and other emerging economies has been at the heart ofmany discussions about globalization. Increasingly, the full scale of the reorga-nization of production across national borders has become apparent, as havethe many different ways in which this takes shape. Since the 1980s, when thefirst formal theories of multinational activity were developed, empirical supporthas been found for the traditional explanations for multinational activity.3 Thereis evidence that multinationals indeed relocate production to save transporta-tion costs and to gain direct access to large markets. Multinationals have alsobeen found to open up affiliates in order to jump tariffs or to move parts oftheir production to where resources are cheap. However, the work on geographic

3 Helpman (1984) is a key theoretical paper on vertical integration that has labour-intensive partsof production relocate to low-wage countries; Brainard (1997) and Markusen (1984) emphasizehorizontal integration and distance for which exports and affiliate production are substitutes.Carr, Markusen, and Maskus (2001) provides empirical evidence of horizontal multinationalactivity. Yeaple (2003) and Hanson, Mataloni, and Slaughter (2005) are two papers thatdocument vertical links. More detailed surveys of the literature are found in Navaretti andVenables (2004) and Markusen (2002).

524 P. Debaere, J.H. Lee, and M. Paik

agglomeration that emphasizes how a multinational’s location decision dependson the decisions of other multinationals addresses the concern that the traditionalfactors do not exhaustively explain the full range of multinational activities andthe allocation decisions that firms make.4

Firms are known to cluster in one geographic area (see Porter 1998). SinceAlfred Marshall, it has been pointed out that there could be different types ofexternal economies to rationalize the geographic clustering of firms. Clusteringmay bring about knowledge and technology spillovers, the increasing availabilityof specialized labour and a growing pool of specialized input providers. Agglom-eration also takes place when firms invest abroad. Foreign investors are morelikely to choose locations where there are many local firms, since their presencemay suggest the mentioned external economies. Interestingly, however, in thecase of foreign direct investment, investors often agglomerate around investorsfrom the same country of origin. In their study of the location decision of U.S.firms in Ireland, Barry, Gorg, and Strobl (2003) argue that investors may exhibita tendency to imitate each other’s location choice due to uncertainty. Since for-eign investors face greater uncertainty than local firms in the host country, theymay interpret the presence of firms from their home country as a positive signalof the location’s attractiveness. It is this characteristic of foreign investment, inparticular, that we investigate.

In the empirical literature, agglomeration is typically interpreted as a positiverelation between a measure of the number of companies in a particular locationand the probability that investors choose that location, which is why the particu-lar path of history matters and why there is persistence in location decisions. Wehypothesize that the presence of more downstream or upstream establishmentsmakes it more likely that investors choose a particular location, which is quiteintuitive. It suggests less costly access to suppliers and buyers. Moreover, the hy-pothesis could be consistent with earlier findings by Belderbos and Carree (2002)that smaller firms tend to follow larger firms. In particular, our hypothesis pro-vides the dependence on inputs from other firms as the reason why this might bethe case. Needless to say, building clusters has been an integral part of the strategyto attract FDI in countries such as Ireland or Costa Rica (Larrain, Lopez-Calva,and Rodriguez-Clare 2001). Note, however, that the policy implications for at-tracting FDI may be different when the presence of upstream or downstreamestablishments is important, irrespective of their nationality, versus when theseestablishments have to be of the same nationality. If the latter holds, attractingsome multinationals from a particular country may help attract others, so theremay be a payoff to bilateral strategies that specifically target certain countries.At the same time, if linkages along national lines matter, potential spillover gainscould be internalized by affiliates of a particular country, and attracting foreignfirms, say, in order to have domestic firms benefit, may be a lot harder.

4 A recent survey by Blonigen (2005) emphasizes the need to go beyond the traditional theories ofthe multinational.


To study agglomeration and its upstream and downstream dimension, we fo-cus on South Korean multinationals and their initial investments in China. Thereare two reasons for focusing on China. The first motivation is entirely pragmatic.South Korean outward FDI is a recent phenomenon and is still relatively limited(especially when compared with outward FDI flows from the U.S. or Japan).Since China attracts most of these new affiliates across regions, and since it hasa sufficient number of regions, we have enough observations to investigate theregional allocation decisions. Note, however, that the pattern of agglomerationthat we observe in China seems to follow a similar pattern in the U.S., for whichwe have far fewer observations.5 The second reason why we focus on China relatesto its being, together with the United States, the highest receiver of the world’sFDI. For this reason it is already important to better understand the locationdecisions in China. Moreover, FDI into China has become one of the premiertopics of policy debates in the region. It has fuelled anxieties of the ‘hollowingout of Korea’s production base as a result of the rush into China,’ as the SouthKorean investment promotion agency KOTRA puts it (see Economist, 25 Au-gust 2001). In this respect, it may be particularly relevant that South Koreanaffiliates abroad are increasingly active in industries such as machinery, trans-portation, and electronics. These industries use a wide range of intermediategoods. In addition, they have been associated with international fragmentationof production (see Hanson, Mataloni, and Slaughter 2005). Therefore, the avail-ability of linkages may be particularly important for these sectors in the wake ofthe Asian currency crisis and the increased liberalization of outward FDI in itswake.

We use a relatively new data set of South Korean FDI in China for theempirical analysis. Different from other data sets, ours is not limited in time spanor scope, which is, of course, related to the fact that China only opened up toSouth Korean FDI fairly recently. The data cover all South Korean investmentin China between 1988 and 2004. The advantage of studying the distributionof FDI within a country rather than across countries is that country-specificfactors can be taken as given. In addition, we can study firm location at aless aggregate level, which is particularly relevant for agglomeration issues. Thechallenge of empirical research on agglomeration is then to properly control foralternative explanations that may explain the presence of clusters of affiliates suchas comparative advantage or government incentives to attract foreign investors.In our preferred specification, we include region-time-specific effects. We alsoinclude a specification with region-specific effects in addition to wages, a region’smarket potential, a measure for regional skill quality as well as controls forChina’s policy initiatives to attract FDI such as the foreign trade zones that havebeen created with the explicit objective of attracting FDI.

5 For a limited set of observations we have evidence from South Korean affiliates across U.S.states that confirms regional clustering along national lines. Also, especially forward linkagesseem to matter for location decisions of South Korean multinationals.


Our study most closely relates to the work of Head, Ries, and Swenson (1995)who, together with Wheeler and Mody (1992), were among the first to studyagglomeration effects for FDI. In particular, Head et al. (1995) examined thelocation of Japanese manufacturing investment across the U.S. states in the 1980s.They also use conditional logit estimates that are well suited for an investigationinto how the variation in location (state) attributes affects the probability thata multinational will choose to set up an affiliate in a particular state. Moreover,Head et al. (1995)’s specific analysis of agglomeration externalities within verticalKeiretsu groupings for Japanese investment in the U.S. paves the way for ourmore general analysis of forward and backward linkages that probes whether theinfluence of linkages extends beyond national lines.6 As a matter of fact, we alsoinvestigate if our results are sensitive to whether firms are part of larger Chaebols.

We first discuss the approach in the next section before we turn to the datathat we use in section 3. In the last two sections, we explain the results and stateour conclusions.

2. Empirical implementation

Conditional logits with the particular place that is chosen by the investor asdependent variables offer a straightforward way to implement location choicemodels. They allow us to investigate how the characteristics of the various loca-tions affect the likelihood of investors investing in a particular place at the timeof the first investment. We follow Head, Ries, and Swenson (1995) who build onMcFadden (1974)’s result that logit choice probabilities can be derived from indi-vidual firm maximization decisions. In particular, the place that offers the highestexpected profitability is chosen as destination. When the production function ofthe affiliate in a particular place is assumed Cobb-Douglas, agglomeration exter-nalities from other companies in the place, together with other production inputs,will affect a firm’s output and profitability in a multiplicative way. In this case,the expected profitability of an affiliate j in place p,� jp, is a log-linear functionof the agglomeration measures and other attributes of the places, all of whichcaptured by the vector X jp. (We drop the time-subscripts for simplicity.)

� jp = β ′X jp + ε jp. (1)

6 Head, Ries, and Swenson (1999) extend their previous analysis as they explicitly control formore factors that characterize the different regions – a key concern in conditional logit analysis.Blonigen, Ellis, and Fausten (2005) find agglomeration effects across both horizontal andvertical Keiretsu groupings. However, their analysis of Japanese multinationals is acrosscountries, rather than states.


If ε jp is Type-I Extreme Value random error, following McFadden (1974), theprobability that j invests in place p equals the following expression:

pr(1j = p) = exp(β ′X jp)∑p

exp(β ′X jp). (2)

The most common formulation of equation (1) is as follows:

πp = θp + α ln Asp + β ln Zp + εp, (3)

where θp represents place-specific effects, Asp stands for agglomeration external-

ities in sector s and place p, and Zp represents other attributes of the differentplaces. It could be argued that the geographic borders of provinces are arbitrary.Therefore, we construct a distance-weighted agglomeration variable for eachprovince that also includes the agglomeration variables of the other provincesweighted by their relative distance.

WAsp = As

p +P∑

l �=p

⎛⎜⎜⎜⎜⎜⎝

Asl ∗ Dist−1

lp

P∑l �=p

Dist−1lp

⎞⎟⎟⎟⎟⎟⎠

, (4)

where Distlp is the distance between capital cities. Hence, WAsp will be higher

where there are many firms nearby.As indicated, to capture agglomeration, we consider WAs

p from two differentangles. On the one hand, we take it to be the number of Korean affiliates withinan industry. On the other hand, we measure the total number of companies in anindustry irrespective of nationality (including local Chinese companies) that arealready active at the time that an investment decision is made.7 Including bothmeasures in the conditional logit will allow us to see which of the two types ofagglomeration has most traction.

Place-specific effects are captured by place-specific dummies that control fortime-invariant factors. These factors capture the geography, the proximity toSouth Korea, the infrastructure, or the presence of a South Korean expatriatecommunity of which all make a place more or less attractive to investors and maybe hard to measure. In addition, we include economic and demographic variablessuch as a place’s education levels, and its average wage rates that vary with thetime of investment. These variables are known determinants of multinational

7 Because of a change in the Chinese data collection (see below), we cannot simply subtract thenumber of South Korean firms from the Chinese firms. However, since the number of SouthKorean establishments is small compared with the total number of companies in an industry,the industry agglomeration variable corresponds roughly to the non-South Korean companies.


activity. Larger markets tend to attract more (horizontal) FDI. Lower wagesmay be attractive for (vertical) FDI that takes advantage of low productioncosts to relocate parts of the production process that used to take place in theSouth Korean parent. We control for the variation in efforts to attract foreigndirect investment by including the number of economic zones in the place. Sincethere is an issue about whether one can appropriately capture all characteristicsthat vary over time and place, our preferred specification includes place-timedummies. Needless to say, if there exists no agglomeration externality and ifall relevant factors that distinguish places are controlled for, the α coefficientsshould be zero.

As indicated, we go beyond this baseline specification for agglomeration. Nextto the regular agglomeration effect, we consider backward and forward linkages,in order to capture the impact of increasing numbers of upstream suppliers ofintermediate goods and downstream buyers of such goods. To generate thesemeasures of forward and backward linkages, we use industry input-output tablesand combine them with the number of companies in a particular place/industry.For the linkages with South Korean companies, we use the South Korean input-output tables and combine them with the number of South Korean affiliatesacross the industries in the Chinese regions. For the linkages at the industry levelirrespective of nationality, we use the Chinese input-output tables combined withthe total number of companies in a particular region/industry that are mostly ofnon-South Korean nationality.8 In each case we capture the strength of forwardlinkages as follows: FLm

p = ∑n δmnAn

p, where δmn is the proportion of sectorm output supplied to sector n and

∑n δmn = 1.9 Again, to take into account

spatial aspects, we construct distance-weighted forward linkages variables in thefollowing way:

WFLmp = FLm

p +∑

n

δmn

P∑l �=p

⎛⎜⎜⎜⎜⎜⎝

Asl ∗ Dist−1

lp

P∑l �=p

Dist−1lp

⎞⎟⎟⎟⎟⎟⎠

. (5)

Hence, WFL will be higher in any place where many downstream firms arealready located nearby. The variable for backward linkages is analogously rep-resented by BLn

p = ∑m γmnAm

p , where γmn is the proportion of sector m output

8 We assume that the link with Korean establishments are reflected in the Korean IO table, whilethat with non-Korean ones are reflected in the Chinese IO table. As a robustness check, however,we use either the Korean or the Chinese IO table for both cases and the results do not depend onit.

9 Consumer demand can be a factor in considering forward linkages. It is not straightforward toconstruct such a variable, however, because consumer demand will be expressed in monetaryterms, while the current linkage variable is based on the number of establishments. Therefore, wewill include market potential as a control variable in some of the specifications.


supplied to sector n and∑

m γmn = 1, and distance-weighted backward linkages,WBL, will be higher where many upstream establishments are already locatednearby. As indicated, we will construct these upstream and downstream linkagevariables specifically for South Korean affiliates as well as for the total number ofcompanies in an industry irrespective of their nationality. Note that the variablesdiffer by place/sector, since the usage of intermediate goods varies by industry.Rewriting the profitability equation (3), we obtain equation (6).

πp = θp +∑

i={SK,I}αia ln WAs

i p +∑

i={SK,I}αi f ln WFLs

i p

+∑

i={SK,I}αi b ln WBLs

i p + β ln Zp + εp.(6)

The new coefficients αf and αb should be significantly positive if an investorchooses a particular place because it has more South Korean (SK) upstream(downstream) affiliates for an industry or, more generally, because the total num-ber of upstream (downstream) companies in this industry is higher irrespectiveof nationality (I). Note that we purposefully include the linkages as well as theregular agglomeration variables for the South Korean affiliates as well as for allfirms irrespective of their destination. In this way, we can see whether the linkageeffects that specify why agglomeration should matter add something to the reg-ular agglomeration effects. Similarly, we should be able to figure out whether theSouth Korean linkages are more or less important than the aggregate industrylinkages.

3. Data and FDI from South Korea

The data of South Korean foreign affiliates are collected by the Export-ImportBank of Korea, which covers the full list of South Korean affiliates establishedworldwide. Relevant for our analysis are the first-time investments of multina-tionals in Chinese regions, which started in 1988. Figures 3 and 4 illustrate thedominant trends of South Korean outward FDI into China in terms of theamount invested and the number of newly established affiliates. Before the mid-1990s FDI gradually increased. Since then, there has been a significant outflow.The late 1990s were the only exception. At the time, South Korea was caughtin the Asian financial crisis. As for outward FDI going into China, the datashow a significant increase in the number of affiliates established as well as in theamount of FDI since 1988. In particular, there was a dramatic increase in FDImoving into China around 1992 when Korea and China entered into diplomaticrelations. Note that there may be some concern that the investments prior to 1992were not merely a function of economic interests. Finally, as is clearly shown infigures 3 and 4, a large percentage of foreign affiliates from South Korea are


02

00

04

00

06

00

0A

mo

un

t(m

il o

f $

)

1980 1985 1990 1995 2000 2005year

World China

FIGURE 3 Amount of South Korean investment to the world and ChinaSOURCE: Export-Import Bank of Korea

05

00

10

00

15

00

# o

f n

ew

aff

ilia

tes

1980 1985 1990 1995 2000 2005year

World China

FIGURE 4 Number of new South Korean affiliates in the world and China for manufacturingSOURCE: Export-Import Bank of Korea


0.2

.4.6

.8China World

Primary Manufacturing

Service

% s

ecto

ral dis

trib

ution

Graphs by dvdp

FIGURE 5 Sectoral shares of South Korea’s new affiliates in China vs. the world as a whole, inpercentagesSOURCE: Export-Import Bank of Korea

located in China. As of 2004, more than 50% of its new affiliates are establishedin China.10

Figures 5 and 6 present the industry characteristics of FDI into China in termsof the number of affiliates. According to figure 5, more than 80% of the affiliatesover the entire period are active in manufacturing in China, which is significantlyhigher than the worldwide share (a bit above 60%). This is in line with theperception of China as the world’s factory. We also isolate the share of affiliatesthat are active in machinery, transportation, and electronics. Not only do thesesectors provide and use many intermediate goods from other sectors. In addition,they are sometimes identified with international fragmentation of production.11

Of interest is to see that the share of the affiliates in these industries has increasedsignificantly after it had been relatively stable before 1999. This could contributeto the importance of linkages.

China has 22 provinces (excluding Taiwan), 5 autonomous regions (Ti-bet, Xinjiang, Inner Mongolia, Ningxia, and Guangxi), and 4 municipalities

10 If we compare the data for China and the U.S., we find that while the volume of FDI to the U.S.is significant and comparable to that of China in later years, the flows go to far more affiliates inChina compared with the U.S., which is why China is better suited for a study of regionalclustering compared with the U.S.

11 Hanson, Mataloni, and Slaughter (2005) find strong vertical FDI activity in industries such asmachinery, transportation, and electronics.


.1.2

.3.4

.5

Perc

enta

ge o

f K

ore

an a

ffili

ate

s

1990 1995 2000 2005year

FIGURE 6 The share of affiliates in machinery, transportation, and electronics in manufacturingSOURCE: Export-Import Bank of Korea

(Beijing, Tianjin, Shanghai, and Chongqing). As figures 1 and 2 show, the SouthKorean affiliates in China are quite concentrated and more concentrated thanthe distribution of the total number of affiliates of any nationality. Four regionsaccount for about 75% of the total population of South Korean affiliates. Theinvestments in Guangdong and Jiangsu, for example, amount to more than 40%of worldwide investment, while those investments account for less than 12%for Korea. Shandong is the premier destination for Korean investment. Inter-estingly, Liaoning, Jilin, and Heilongjiang seem to be attractive locations forKorean firms, but not for those of other countries. This should perhaps not be somuch of a surprise, since many Korean-Chinese who speak Korean live in theseprovinces, which are adjacent to North Korea.12 Note that this concentration ofSouth Korean affiliates is also found in the U.S., the other large country in thedata set of which we have a regional distribution. Here again, we find a higherconcentration among South Korean affiliates than for all the affiliates combined.Almost 60% of Korean FDI is concentrated in California, which is five timeshigher than the world average.13

12 As we will not use worldwide FDI in our conditional logits, but rather the number of companiesfrom other nationalities in a particular region/industry irrespective of their nationality, we wantto note that the correlation between the regional distribution of worldwide FDI and that ofoverall manufacturing output is, as one would expect, fairly high at 74%.

13 When we construct a Herfindahl index to measure the concentration of affiliates across states,the pattern for China and the U.S. is comparable. There is a higher concentration for South


We end up working with 25 regions. Since the five autonomous regions aregeographically separate entities and the investments are reported together, wecannot include them in the analysis.14 At the same time, we merge the data forthe municipality of Chongqing with Sichuan Province, since it was separatedonly in 1997.

We measure agglomeration using the number of Korean affiliates alreadyactive in the region in the same manufacturing industry in the year before aninvestment takes place. The Export-Import Bank of Korea records South Ko-rean outward FDI according to a total of 69 manufacturing industries, whichis similar to the Korea Standard Industry Classification (KSIC) 4-digit level.Accordingly, the agglomeration variable for South Korean affiliates is specifiedfor 69 industries. For the total number of companies at the industry level thatare overwhelmingly of non-South Korean nationality, we use the data from theChina Industry Statistics Yearbook series. The data contain the total number offirms and output according to 18 manufacturing industries. In one specification,we will use both (different) classifications. In another, we will match as closelyas we can the Chinese classification with the South Korean data by aggregatingKorean data up to 18 (roughly) comparable industries.15

When we do the estimation, we will first focus on the period since 1999, beforewe turn to the entire period after 1992. There are two reasons for doing so. Theperiod since the Asian financial crisis has seen the strongest increase in affiliatesfrom electronics, materials, and transportation that are especially relevant forour analysis. More important, a data issue complicates the analysis for the entireperiod. The China statistics department changed the data collection classificationbetween 1997 and 1998. Owing to that change, the total number of companies in1998 dropped to one-third of the 1997 data across all the industries. To controlfor this anomaly, we will interact the aggregation variables with a dummy forthe post-1998 years in our analysis when we focus on the entire period since1992. Finally, to avoid missing values in a log transformation, we add one to thisvariable as previous studies have done.16

When we focus on the supply and demand of intermediate goods by SouthKorean affiliates, we use the Korean input-output table from the year 2000 tomeasure firm linkages.17 To merge input-output tables and Korean FDI data, we

Korean vs. total foreign affiliates across states in both cases: the Herfindahl indices for SouthKorean affiliates in China and in the U.S. are respectively 0.182 and 0.338; for total foreignaffiliates we find 0.129 in China and 0.043. in the U.S.

14 The investment into those regions is less than 10.15 The 18 industries are general industry machinery, other machinery, non-metal mineral, textile,

synthetic fibre, food, grain-mill products, beverages, instruments, automobile, electronic andelectrical machinery, electronic and communication components, primary metal, fabricatedmetal products, printing and allied products, coke and petroleum, chemical and drug.

16 An easy way to rationalize this is to argue that the investing firm does take its own presence inthe region into account as it decides whether to invest in a region. We follow Head, Ries, andSwenson (1995).

17 Since the industry shares do not change much during the sample period, the 2000 table is usedfor all years. Alternatively, one could use the IO tables that were published with five-year


have to concord the industry classifications, as there is a slight variation betweenboth sources.18 For example, the input-output table specifies semiconductors(KSIC 3211) and other related devices (KSIC 3219) separately, while in the FDIdata set, both industries are classified as semiconductors and related devices. Inthis case, we combine KSIC 3211 and KSIC 3219 to match the FDI classification.When some industries are more finely defined in the FDI data set than in theinput-output table, we adjust the sectors accordingly. In the end, our adjustedinput-output table consists of 53 industries. In order to be able to construct thesecond agglomeration variable at the industry level irrespective of nationality,we rely on the Chinese input-output table 1995, which comprises 15 industries.19

Finally, in order to construct the linkage variables as described in section 2, weinteracted the Korean input-output table with the number of Korean affiliatesand the Chinese input-output table with the number of firms of all nationalitiesand also add one to them for the same reason we do with the agglomerationvariable. While it is not feasible to exactly match the Chinese and South Koreaninput-output tables, we will also present estimates with a more aggregate SouthKorean input output table of 18 sectors.

We control for regional economic and demographic factors in the estimationin two ways. In our preferred specification, we include time-region dummies.We also show results for an alternative specification that includes some of theknown determinants of FDI. It is well known, for example, that a larger re-gional economy attracts more FDI. At the same time, as Head and Mayer (2004)suggest, market potential is an important factor that may affect an investor’slocation decision. To control for market potential and a region’s size, we includethe distance-weighted real GDPs of all regions. We take the real GDP data fromvarious issues of the China Statistical Yearbook. The distance between regionsis measured as the distance between the provincial capitals, which is taken fromyahoo.com. We control for the labour costs by including the average level of re-gional staff and worker wages from the China Statistical Yearbook. Lower wagerates could be more attractive to investors in search of cheap labour. Furthermore,to control for the quality of workers, we include the ratio of high school grad-uates to the total population. We compute this ratio from the China StatisticalYearbook. Finally, we also consider a variable that reflects the government’s rolein attracting FDI. We choose the number of the special economic zones (SEZ)in a region that was especially created for foreign companies.20 There are manydifferent types of economic zones such as open coastal cities (OCCs), economicand technological development zones (ETDZs), open coastal areas (OCAs),

intervals. A drawback of going that route is that the classification also changes over the timeperiod.

18 The input-output table is published by the Bank of Korea, while the FDI data come from theExport-Import Bank.

19 We also used the table published by the World Bank, which is based on GTAP 4 database.Input-output coefficients are fairly stable and analysis using either table produced the similarresults.

20 Cheng and Kwan (2000) provide evidence of the significant role of special economic policy.


TABLE 1Summary statistics of regressors

Variable Mean St. Dev. Min Max

Agglomeration 6.618 0.819 3.209 8.878Agg. by SK affiliates 1.821 1.153 0 5.77Forward linkages 7.867 0.636 6.254 9.447Backward linkages 8.000 0.62 6.335 9.818For. link. by SK affiliates 1.644 0.947 0 5.28Back. link. by SK affiliates 1.517 0.883 0.003 4.92

SOURCE: Korean data are from Export-Import Bank of Korea and Bank of Korea. Non-Koreandata in China are from Chinese Statistical Yearbook and China Industry Statistical Yearbook. Allvariables are in log term.

technology industry development zones (TIDZs), bonded zones (BZs), bordereconomic cooperation zones (BECZs), and export processing zones (EPZs). Asof 2004, there was a lot of variation in the total number of zones in a region, withGuangdong having as many as 20 economic zones, for example.

Table 1 reports the summary statistics for the main variables of our analy-sis after taking into account spatial aspects. Agglomeration measures the totalnumber of firms in an industry, irrespective of their nationality. Agglomerationby SK affiliates counts only the South Korean affiliates in an industry. Similarly,we have forward and backward linkages involving all the firms in an industry aswell as those involving only the South Korean affiliates.

4. Results

Table 2 reports the estimation results for the period after the Asian financial crisis.As mentioned before, this period saw an increase in the share of affiliates fromsectors prone to vertical integration and keen on international fragmentation.At the same time, as far as the total number of firms goes, the data are the mostconsistent for this period. The results present the estimates of equation (6) thatspecify the factors that determine an investor’s decision to invest in a particularChinese region. In all columns, except for the fourth one, the equation includestime-region-specific effects. The first two columns present familiar estimates ofequation (6) that have been used in previous studies. The estimates in the firstcolumn suggest that firms agglomerate by industry regardless of nationalities.Those in the second column indicate that the decision to invest in a particularChinese region is determined not only by the agglomeration of companies of anynationality in a given industry, but also in particular by the number of SouthKorean affiliates in that industry. The coefficients are significantly positive andin the range of previous studies. The likelihood ratio test between the specifica-tion in column (1) vs. (2) overwhelmingly rejects the hypothesis that the South


TABLE 2Estimation after 1999

(6)(5) Non-

(1) (2) (3) (4) Aggregation Chaebol

Agglomeration 1.061 0.712 0.46 0.53 0.445 0.486[0.070]∗∗∗ [0.073]∗∗∗ [0.172]∗∗∗ [0.167]∗∗∗ [0.177]∗∗ [0.183]∗∗∗

Agg. by SK affiliates 0.766 0.348 0.315 0.571 0.345[0.049]∗∗∗ [0.087]∗∗∗ [0.086]∗∗∗ [0.147]∗∗∗ [0.091]∗∗∗

Forward linkages 0.162 0.214 0.283 0.121[0.366] [0.362] [0.305] [0.383]

Backward linkages 0.066 −0.261 −0.271 0.05[0.424] [0.391] [0.354] [0.443]

For. link. by SK affiliates 0.469 0.563 0.501 0.492[0.140]∗∗∗ [0.135]∗∗∗ [0.243]∗∗ [0.145]∗∗∗

Back. link. by SK affiliates 0.542 0.55 0.206 0.506[0.134]∗∗∗ [0.130]∗∗∗ [0.244] [0.139]∗∗∗

Market potential 3.491[3.738]

High school graduates 0.099[0.610]

Wage rate 6.946[1.240]∗∗∗

Economic zones 1.649[0.638]∗∗∗

Province dummy no no no yes no noProvince & year dummy yes yes yes no yes yesNo. of choices 25 25 25 25 25 25No. of investors 4264 4264 4264 4264 4264 4028Pseudo-R2 0.37 0.38 0.38 0.37 0.37 0.39Log-likelihood −8686.2 −8561.67 −8543.01 −8651.49 −8599.23 −7964.03

SOURCE: Korean data are from Export-Import Bank of Korea and Bank of Korea. Non-Koreandata are from Chinese Statistical Yearbook and China Industry Statistical Yearbook. Distance is fromyahoo.com. All variables except for dummies are in log term. Standard errors are in parentheses.∗significant at 10%, ∗∗significant at 5%, ∗∗∗significant at 1%.

Korean agglomeration variables have no explanatory power.21 Note that the in-terpretation of the coefficient estimate as the average probability elasticity needssome care in a conditional logit model. It can be shown that the average prob-ability of how any regressor impacts the location choice over all choosers andlocation choices should be calculated as follows: (S − 1)/S times the regressor’sestimated coefficient, where S is the number of location choices (see Head, Ries,and Swenson 1995). Since there are 25 locations in our study, our estimates showthat a 10% increase in the distance-weighted number of Korean affiliates in oneregion will increase the probability that investors choose that region by around10% (0.96 × 1.061 × 10).

21 The test statistic is 249 and the critical value at 0.005 level is 8.


As argued before, there are various ways to interpret these findings, and itis not clear why firms would agglomerate in a particular location. We thereforeadd to the regular agglomeration effect the effect due to forward and back-ward linkages. Moreover, we consider both the agglomeration linkages for SouthKorean affiliates and the agglomeration linkages at the industry level for compa-nies of any nationality (including local Chinese companies). In the third columnof table 2, we include the linkage variables that are described in section 2. Theregular agglomeration externalities remain strongly positive. The result showsthat both forward and backward linkage effects are significant for South Koreanestablishments. The likelihood ratio test prefers the specification in column (3)that includes the South Korean linkage variables over that in column (2).22 As ex-pected, the magnitude of the regular agglomeration effect decreases significantlyas forward and backward linkages are included. Interestingly enough, however,the linkage effects at the industry level across nationalities are not significant.This suggests that, while the presence of many companies in an industry mat-ters, the nationality of the establishments is key for the specific upstream anddownstream links that are directly aligned with the specific production process.

In the fourth column of table 2, we drop the time-region effects and includeregion-specific dummies together with other more traditional determinants ofindustry location that vary over time. As noted, all our measures of marketpotential, education, and the wage and economic policies meant to attract multi-nationals enter with a positive sign. However, only the wage and the number ofeconomic zones are statistically significant. The positive coefficient on the wageruns counter to our initial intuition that multinationals might seek low-wageregions. This may suggest that the wage also picks up the quality/education levelof the labour force.

As mentioned, there is an issue about the different classification of the in-dustries according to the Chinese versus the South Korean statistics. The fifthcolumn reports estimates when we (imperfectly) map the South Korean indus-try classification into its Chinese counterpart and use a more aggregate SouthKorean input output table. For the most part, the results hold up. Only thebackward South Korean link loses significance, while maintaining the same sign.

Finally, we focus on the role of Chaebol for our results. It is well known thatChaebol, South Korea’s conglomerates, play a prominent role in South Korea’sindustrial texture. We want to investigate whether our results are driven by theselarge corporations. It turns out that they are not. When we drop all the largercorporations from the sample, all variables of interest retain the same signs andsignificance.23 These findings then add an interesting dimension to the existingliterature. As Belderbos and Carree (2002) have noted, small firms are followers.What our results show is that smaller firms are followers, especially becausethey go where suppliers of inputs and buyers of their intermediate goods are

22 The test statistic is 37 and the critical value at 0.005 is 15.23 In the survey, the investing firms declare themselves large or not. Some 8% is large.


0.2

.4.6

.81

Pe

rce

nta

ge

of

Ko

rea

n a

ffili

ate

s

1985 1990 1995 2000 2005year

FIGURE 7 Fraction of affiliates from large multinationals over timeSOURCE: Export-Import Bank of Korea

more plentiful. Moreover, figure 7 shows that the fraction of big firms in totalfirms has been decreasing over time, suggesting in addition that initially largermultinationals went abroad.

In table 3, we extend the sample. We present estimates for the post-1992 period.We estimate from 1992 onward rather than 1988, since there is some concern thatthe location choice before the diplomatic relations between China and SouthKorea were initiated in 1992 may not have been purely for economic reasons.24

Note that we included early establishments going back to 1988 in the count offirms to construct the agglomeration variables when we estimated the locationdecision after 1992. The makeup of table 3 mimics that of the previous table forthe time since 1999, and the results largely correspond to those of the more recentperiod.25 The main difference is that the estimates are somewhat weaker and lessprecisely estimated.

The last issue that we address is the robustness of the results. In particu-lar, we investigate the independence of irrelevant alternatives (IIA) assumption

24 It turns out that estimates for the entire sample, starting from 1988 are qualitatively the same,but somewhat weaker. Alternatively, one could argue that the early years of the data set are notof major interest to uncover agglomeration effects, since there were no South Korean affiliates in1988.

25 The likelihood ratio tests comparing the specification in column (7) vs. (8) and (8) vs. (9) prefercolumns (8) and (9).


TABLE 3Estimation after 1992

(12)(11) Non-

(7) (8) (9) (10) Aggregation Chaebol

Agglomeration 0.879 0.655 0.481 0.219 0.44 0.587[0.097]∗∗∗ [0.097]∗∗∗ [0.160]∗∗∗ [0.132]∗ [0.160]∗∗∗ [0.172]∗∗∗

Agg. ∗D(year>= 1998) 0.18 0.066 0.046 0.405 0.091 0.017[0.119] [0.119] [0.119] [0.085]∗∗∗ [0.118] [0.128]

Agg. by SK affiliates 0.743 0.418 0.507 0.458 0.403[0.038]∗∗∗ [0.067]∗∗∗ [0.064]∗∗∗ [0.106]∗∗∗ [0.070]∗∗∗

Forward linkages 0.101 0.05 0.085 −0.162[0.304] [0.297] [0.264] [0.322]

Backward linkages 0.082 0.305 −0.011 0.258[0.353] [0.269] [0.293] [0.373]

For. link. by SK affiliates 0.338 0.205 0.326 0.357[0.105]∗∗∗ [0.097]∗∗ [0.178]∗ [0.111]∗∗∗

Back. link. by SK 0.457 0.3 0.316 0.447affiliates [0.101]∗∗∗ [0.093]∗∗∗ [0.189]∗ [0.106]∗∗∗

Market potential 1.247[0.626]∗∗

High school graduates −0.105[0.226]

Wage rate 0.074[0.330]

Economic zones 0.395[0.148]∗∗∗

Province dummy no no no yes no noProvince & year dummy yes yes yes no yes yesNo. of choices 25 25 25 25 25 25No. of investors 6863 6863 6863 6863 6863 6307Pseudo-R2 0.35 0.36 0.36 0.35 0.36 0.37Log-likelihood −14260.2 −14061.86 −14042.03 −14319.27 −14145.29 −12715.11

SOURCE: Korean data are from Export-Import Bank of Korea and Bank of Korea. Non-Koreandata are from Chinese Statistical Yearbook and China Industry Statistical Yearbook. Distance is fromyahoo.com. All variables except for dummies are in log term. Standard errors are in parentheses.∗significant at 10%, ∗∗significant at 5%, ∗∗∗significant at 1%.

that is implied in a conditional logit analysis. In a conditional logit, the relativeprobability of choosing between two alternatives should not depend on the avail-ability of a third alternative. The IIA assumption hinges upon the identical andindependent error terms. As argued by Head, Ries, and Swenson (1995), the in-clusion of alternative specific constants (in our case, regional dummies) allows forconditional logits in the presence of violations of IIA, as long as investors haveuniform perceptions about the substitutability between states. At the same time,the regional dummies complicate formal testing of IIA, since they yield differentnumbers of parameters across specifications. We therefore compare the estimatesof the critical variables of interest, as we exclude several regions with the baselineestimates for the full set of choices. When the coefficients and significance levels


are relatively stable, we regard the IIA assumption as valid, as is done in the pre-vious studies. We exclude Shandong, three Northeast provinces (Jilin, Liaoning,and Heilongjiang), and three municipalities (Beijing, Tianjin, and Shanghai) inturn in the second and third column of table 4. Note that the first column foreach time period is the standard estimate (4) from table 2 and (10) from table 3,which include regional variables with region-specific dummies. For reference,Shandong is the location of most Korean firms; the Northeast provinces are at-tractive regions for Korean investors, since there are many Korean Chinese; andthe municipality itself has economic significance. For the period after 1999, thecoefficients and significance levels are relatively stable. For the entire period after1992 we have comparable results that, as before, tend to be somewhat smallerand somewhat less precisely estimated.

5. Conclusion

China, together with the United States, has topped the list of recipients of FDIfor a number of years. This has heightened the interest in gaining a better under-standing of what drives multinationals into China and what explains the localdispersion of multinationals as China is rapidly being integrated in the worldeconomy. With an unpublished data source for all of Korea’s affiliates acrossChina’s regions, we investigate whether and how agglomeration affects the allo-cation decision along industry lines and along national lines. In particular, weextend the usual agglomeration analysis and investigate next to the regular ag-glomeration effect the impact of backward and forward linkages. Since the workby Hirschman (1958), there has been an active interest in forward and backwardlinkages. Moreover, the idea of forming clusters of economic activity has beencentral in the attempts to attract FDI in countries such as Ireland and CostaRica.

We find that forward and backward linkages interacted with the presenceof other Korean affiliates in China play a significant role in determining thelocation of Korean FDI in China. At the same time, however, the general forwardand backward linkages at the industry level, irrespective of nationality, do notseem to matter much. These findings are fairly robust and are not driven bythe South Korean Chaebol. Our results imply the presence of multinationalsfrom one country of origin attracts other firms that country. From a policyperspective, this seems to imply a bilateral approach. Indeed, if the objective ofthe Chinese government is to attract more multinationals, it may be worthwhileto target specific countries. At the same time, our findings also imply a note ofcaution. With clustering along national lines, potential spillovers are likely to beinternalized among the affiliates from one and the same country. Consequently,if generating spillovers for the local industry is part of the reason for attractingforeign companies, our results indicate that this may be a difficult objective torealize. Interestingly enough, recent research by Girma and Gong (2008) and

TA

BL

E4

Rob

ustn

ess

1999

–200

419

92–2

004

(13)

(14)

(15)

(16)

(17)

(18)

Exc

ludi

ng(4

)Sh

ando

ngN

orth

east

Mun

icip

alit

ies

(10)

Shan

dong

Nor

thea

stM

unic

ipal

itie

s

Agg

lom

erat

ion

0.53

0.57

50.

758

0.44

30.

219

0.26

60.

331

0.27

6[0

.167

]∗∗∗

[0.1

98]∗∗

∗[0

.188

]∗∗∗

[0.1

99]∗∗

[0.1

32]∗

[0.1

52]∗

[0.1

65]∗∗

[0.1

53]∗

Agg

.by

SKaf

filia

tes

0.31

50.

404

0.27

20.

249

0.50

70.

494

0.50

60.

499

[0.0

86]∗∗

∗[0

.105

]∗∗∗

[0.1

00]∗∗

∗[0

.097

]∗∗[0

.064

]∗∗∗

[0.0

79]∗∗

∗[0

.076

]∗∗∗

[0.0

72]∗∗

∗

For

war

dlin

kage

s0.

214

0.1

0.19

40.

312

0.05

0.20

40.

177

−0.0

22[0

.362

][0

.452

][0

.392

][0

.432

][0

.297

][0

.369

][0

.343

][0

.349

]B

ackw

ard

linka

ges

−0.2

61−0

.127

−0.6

43−0

.246

0.30

50.

118

−0.5

220.

52[0

.391

][0

.458

][0

.420

][0

.468

][0

.269

][0

.320

][0

.338

][0

.317

]F

or.l

ink.

bySK

affil

iate

s0.

563

0.38

20.

623

0.71

70.

205

0.19

70.

318

0.21

6[0

.135

]∗∗∗

[0.1

63]∗∗

[0.1

55]∗∗

∗[0

.158

]∗∗∗

[0.0

97]∗∗

[0.1

19]∗

[0.1

17]∗∗

∗[0

.110

]∗

Bac

k.lin

k.by

SKaf

filia

tes

0.55

0.63

60.

413

0.45

70.

30.

333

0.11

30.

289

[0.1

30]∗∗

∗[0

.168

]∗∗∗

[0.1

55]∗∗

∗[0

.146

]∗∗∗

[0.0

93]∗∗

∗[0

.118

]∗∗∗

[0.1

14]

[0.1

03]∗∗

∗

Pro

vinc

edu

mm

yye

sye

sye

sye

sye

sye

sye

sye

sN

o.of

choi

ces

2524

2222

2524

2222

No.

ofin

vest

ors

4264

2571

3583

3464

6863

4400

5312

5570

Pse

udo-

R2

0.37

0.28

0.43

0.45

0.35

0.28

0.42

0.43

SOU

RC

E:K

orea

nda

taar

efr

omE

xpor

t-Im

port

Ban

kof

Kor

eaan

dB

ank

ofK

orea

.Non

-Kor

ean

data

are

from

Chi

nese

Stat

isti

calY

earb

ook

and

Chi

naIn

dust

rySt

atis

tica

lYea

rboo

k.D

ista

nce

isfr

omya

hoo.

com

.All

vari

able

sexc

eptd

umm

iesa

rein

log

term

.Sta

ndar

der

rors

are

inpa

rent

hese

s.A

llsp

ecifi

cati

ons

incl

ude

mar

ketp

oten

tial

,wag

era

te,t

hera

tio

ofhi

ghsc

hool

grad

uate

s,an

dth

enu

mbe

rofe

cono

mic

zone

s.∗ s

igni

fican

tat1

0%,∗∗

sign

ifica

ntat

5%,∗∗

∗ sig

nific

ant

at1%

.


Girma, Gong, and Gorg (2008) indicates that there have been limited spilloversfrom FDI in China on its state-owned enterprises, which is consistent with ouranalysis.

Appendix

In this appendix, we report the empirical results for the South Korean investmentinto the U.S. from 1999 to 2004 in table A2. First of all, figure A1 compares thenumber of newly established affiliates in the world, China, and the U.S. FigureA2 shows the distribution of multinationals across states. During the period,South Korean investors invest in a total of 41 U.S. states including DC. Tocalculate the total number of U.S. companies at the industry level, we use thedata from the Statistics of U.S. Businesses. The data contain the total numberof firms according to 18 manufacturing industries. The input-output table ofthe U.S. is from the Bureau of Economic Analysis, which also comprises 18sectors. The summary statistics are reported in table A1. The distance betweenregions is measured as the distance between the state capitals, which is takenfrom yahoo.com.

0500

1000

15

00

# o

f n

ew

aff

ilia

tes

1980 1985 1990 1995 2000 2005year

World China

US

FIGURE A1 Number of new South Korean affiliates in the world and the U.S.SOURCE: Export-Import Bank of Korea

0.2.4.6

Percentage of World/Korean affiliates

AlabamaAlaska

ArizonaArkansasCaliforniaColorado

ConnecticutDelaware

District of ColumbiaFlorida

GeorgiaHawaiiIdahoIllinois

IndianaIowa

KansasKentuckyLouisiana

MaineMaryland

MassachusettsMichigan

MinnesotaMississippi

MissouriMontana

NebraskaNevada

New HampshireNew JerseyNew Mexico

New YorkNorth Carolina

North DakotaOhio

OklahomaOregon

PennsylvaniaRhode Island

South CarolinaSouth Dakota

TennesseeTexasUtah

VermontVirginia

WashingtonWest Virginia

WisconsinWyoming

Wo

rld

K

ore

a

FIG

UR

EA

2D

istr

ibut

ion

ofm

ulti

nati

onal

sin

the

U.S

.SO

UR

CE

:For

wor

ldw

ide

mul

tina

tion

alfir

ms:

U.S

.Cen

sus

Bur

eau.

For

Kor

ea:f

rom

Exp

ort-

Impo

rtB

ank

ofK

orea


TABLE A1Summary statistics of regressors (U.S.)

Variable Mean St. Dev. Min Max

Agglomeration 6.228 0.657 4.799 8.353Agg. by SK affiliates 0.468 0.512 0 4.053Forward linkages 6.210 0.77 4.465 9.036Backward linkages 6.471 0.493 5.199 8.288For. link. by SK affiliates 0.412 0.436 0.003 3.704Back. link. by SK affiliates 0.324 0.357 0.042 3.085

SOURCE: Korean data are from Export-Import Bank of Korea and Bank of Korea. Non-Koreandata in the U.S. are from U.S. Census Bureau. All variables are in log term.

TABLE A2Estimation after 1999 (U.S.)

(4)Non-

(1) (2) (3) Chaebol

Agglomeration −0.666 −0.029 0.131 0.516[0.742] [0.735] [0.779] [0.825]

Agg. by SK affiliates 0.912 0.425 0.256[0.132]∗∗∗ [0.250]∗ [0.279]

Forward linkages 0.797 0.692[0.709] [0.757]

Backward linkages −0.144 0.341[1.290] [1.385]

For. link. by SK affiliates 0.801 1.005[0.338]∗∗ [0.378]∗∗∗

Back. link. by SK affiliates −0.211 −0.388[0.453] [0.515]

Province dummy no no no noProvince & year dummy yes yes yes yesNo. of choices 41 41 41 41No. of investors 455 455 455 388Pseudo-R2 0.55 0.56 0.57 0.57Log-likelihood −766.59 −740.97 −734.67 −571.79

NOTES: All variables except for dummies are in log term. Standard errors are in parentheses.SOURCE: Korean data are from Export-Import Bank of Korea and Bank of Korea. Non-Koreandata are from U.S. Census Bureau and Bureau of Economic Analysis. Distance is from yahoo.com.

References

Amiti, Mary, and Beata Smarzynska Javorcik (2008) ‘Trade costs and location of foreignfirms in China,’ Journal of Development Economics 85, 129–49

Barry, Frank, Holger Gorg, and Eric Strobl (2003) ‘Foreign direct investment, agglomera-tion and demonstration effects: an empirical investigation,’ Review of World Economics139, 583–600


Belderbos, Rene, and Martin Carree (2002) ‘The location of Japanese investments inChina: agglomeration effects, Keiretsu, and firm heterogeneity,’ Journal of the Japaneseand International Economics 16, 194–211

Blonigen, Bruce A. (2005) ‘A review of the empirical literature on FDI determinants,’Atlantic Economic Journal 33, 383–403

Blonigen, Bruce A., Christopher J. Ellis, and Dietrich Fausten (2005) ‘Industrial groupingsand foreign direct investment,’ Journal of International Economics 65, 75–91

Brainard, S. Lael (1997) ‘An empirical assessment of the proximity-concentration trade-off between multinational sales and trade,’ American Economic Review 87, 520–44

Carr, David L., James R. Markusen, and Keith E. Maskus (2001) ‘Estimating theknowledge-capital model of the multinational enterprise,’ American Economic Review91, 693–708

Cheng, Leonard K., and Yum K. Kwan (2000) ‘What are the determinants of the lo-cation of foreign direct investment? The Chinese experience,’ Journal of InternationalEconomics 51, 379–400

Girma, Sourafel, and Yundan Gong (2008) ‘FDI, linkages and the efficiency ofstate-owned enterprises in China,’ Journal of Development Studies 44, 728–49

Girma, Sourafel, Yundan Gong, and Holger Gorg (2008) ‘Foreign direct investment,access to finance, and innovation activity in Chinese enterprises,’ World Bank EconomicReview 22, 367–82

Gorg, Holger, and Eric Strobl (2002) ‘Multinational companies and indigenous develop-ment: an empirical analysis,’ European Economic Review 46, 1305–22

Hanson, Gordon H., Raymond J. Mataloni, and Matthew J. Slaughter (2005) ‘Verticalproduction networks in multinational firms,’ Review of Economics and Statistics 87,664–78

Head, Keith, and Thierry Mayer (2004) ‘Market potential and the location of Japaneseinvestment in the European Union,’ Review of Economics and Statistics 86, 959–72

Head, Keith, John Ries, and Deborah Swenson (1995) ‘Agglomeration benefits and loca-tion choice: evidence from Japanese manufacturing investments in the United States,’Journal of International Economics 38, 223–47

— (1999) ‘Attracting foreign manufacturing: investment promotion and agglomeration,’Regional Science and Urban Economics 29, 197–218

Helpman, Elhanan (1984) ‘A simple theory of international trade with multinationalcorporations,’ Journal of Political Economy 92, 451–71

Hirschman, Albert O. (1958) The Strategy of Economic Development (New Haven, CT:Yale University Press)

Larrain, Felipe B., Luis F. Lopez-Calva, and Andres Rodriguez-Clare (2001) ‘Intel: a casestudy of foreign direct investment in Central America,’ in Growth and Internationaliza-tion, ed. Felipe B. Larrain (Cambridge, MA: Harvard University Press)

Markusen, James R. (1984) ‘Multinationals, multi-plant economies, and the gains fromtrade,’ Journal of International Economics 16, 205–26

— (2002) Multinational Firms and the Theory of International Trade (Cambridge, MA:MIT Press)

Markusen, J., and A. Venables (1999) ‘Foreign direct investment as a catalyst for industrialdevelopment,’ European Economic Review 43, 335–56

McFadden, Daniel S. (1974) ‘Conditional logit analysis of qualitative choice behavior,’ inFrontiers in Econometrics, ed. P. Zarembka (New York: Academic Press)

Navaretti, Giorgio Barba, and Anthony J. Venables (2004) Multinational Firms in theWorld Economy (Princeton, NJ: Princeton University Press)


Porter, Micheal (1998) ‘Clusters and the new economics of competition,’ Harvard BusinessReview 76, 77–90

Wheeler, David, and Ashoka Mody (1992) ‘International investment location decisions:the case of U.S. firms,’ Journal of International Economics 33, 57–76

Yeaple, Stephen Ross (2003) ‘The role of skill endowments in the structure of U.S.outward foreign direct investment,’ Review of Economics and Statistics 85, 726–34

agglomeration, backward and forward linkages: evidence from … · agglomeration, backward and...

Documents